Cover the lasym cos part of the vacuum grpmn and bvec

[CharlesCNorton]

Change - covers the cos halves of grpmn and bvec in the NESTOR vacuum tests, on a new asymmetric free-boundary case, and drops the four TODO(jons) markers that stood in for that coverage.

Cause - grpmn and bvec carry a second symmetry index for lasym, but educational_VMEC dumped only the first slice; analyt.f90 notes it at the call as ! missing dim: (ndim: 1 or 2). With no reference for the cos half, the asymmetric vacuum path was untested. jonathanschilling/educational_VMEC#30 writes it out.

Evidence - the reference is a free-boundary lasym = T cth-like case at ns = 15, mpol = 5, ntor = 4, nfp = 5, generated with the mgrid VMEC++ already uses for cth_like_free_bdy_asym, md5-identical, and with every namelist parameter matching cth_like_free_bdy_asym.json. Both halves agree at 1e-9, the same tolerance the symmetric case uses.

Tests - FourPTest, FourIAccumulateGrpmnTest and AnalytTest gain cth_like_free_bdy_asym at iteration 53 and compare grpmn_cos and bvec_cos where lasym is set. Both C++ suites pass, 44/44.

Scope - test-only. The symmetric cases are unchanged and the new comparisons are skipped when lasym is false. Needs educational_VMEC#30 for anyone regenerating the reference; the JSONs are checked in, so the tests do not depend on it.

[jurasic-pf]

These comments are quire hard to read. The code should still be understandable and maintainable by humans, and not just ai.

[CharlesCNorton]

@jurasic-pf Rewriting them. Separately, #821 proposes fixing a layout for PR descriptions in AGENTS.md, for the same reason.

[CharlesCNorton]

@jurasic-pf Rather than reword the markers I removed the reason for them: educational_VMEC now writes the cos half (jonathanschilling/educational_VMEC#30), and the tests compare it on a new asymmetric free-boundary case at 1e-9. Retitled, since the PR no longer records a gap.

[jurasic-pf]

Perfect! With test coverage and easy to read code